Clinical manifestations and response to therapies in ulcerative colitis (UC) are heterogeneous, yet patient classification criteria for tailored therapies are currently lacking. Here, we present an unsupervised molecular classification of UC patients, concordant with response to therapy in independent retrospective cohorts. We show that classical clustering of UC patient tissue transcriptomic data sets does not identify clinically relevant profiles, likely due to associated covariates. To overcome this, we compare cross-sectional human data sets with a newly generated longitudinal transcriptome profile of murine DSS-induced colitis. We show that the majority of colitis risk-associated gene expression peaks during the inflammatory rather than the recovery phase. Moreover, we achieve UC patient clustering into two distinct transcriptomic profiles, differing in neutrophil-related gene activation. Notably, 87% of patients in UC1 cluster are unresponsive to two most widely used biological therapies. These results demonstrate that cross-species comparison enables stratification of patients undistinguishable by other molecular approaches.